Add vp audit, Python SDK (visionpack.sdk), and security/perf hardening - #11
Merged
Merged
Conversation
Security & robustness (from a code audit of the framework): - Sanitize class names used as export path components (formats/base.py safe_path_component): names arriving from imported data (folder names, COCO categories, classes.txt) could contain separators or `..` and write outside the export directory via `vp export --format imagefolder`. - Catch PIL.Image.DecompressionBombError in media probes so one hostile image (header claiming absurd dimensions) becomes a per-file ingest failure instead of aborting the whole import/sync batch. Performance: - New Index.asset_ids() (SELECT id + dirty overlay) replaces materializing every asset record where sync/import only need the id set. vp audit (roadmap Phase B — label-health audit): - visionpack/audit.py: duplicate same-class boxes (IoU), degenerate (tiny) boxes, edge-pinned and whole-image boxes, aspect-ratio outliers, rare classes, class imbalance; thresholds via validation.audit in visionpack.yaml or per-run flags. - CLI: vp audit [--json] [--fail-on-findings] (advisory by default). Python SDK: - visionpack/sdk: VisionPackClient wraps the full lifecycle — init/open, import_dir, sync/plan_sync, validate, audit, stats, splits, snapshots (with read-only checkout(version) views), export, and the model loop (load_predictions, evaluate, autolabel, annotation_queue). Mutating methods take the same project lock as the CLI; returns mirror the --json envelope shapes. Docs: new docs/sdk.md, vp audit section in usage.md, json-output table row, ARCHITECTURE module map + roadmap checkbox, README and CHANGELOG updates. Tests: 22 new (tests/test_audit.py, tests/test_sdk.py); 165 total passing. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RKKNgi6ahA6jYnX2CzDEG9
Three more roadmap items, continuing the previous batch: Distribution drift (Phase B): - visionpack/drift.py: class-distribution drift between two snapshots, computed from the stats frozen at snapshot time — per-class count and distribution-share deltas (biggest movers first) plus smoothed KL and Jensen-Shannon divergence as single thresholdable scores. - CLI: vp diff v1 v2 --drift (human table + a "drift" object in --json). Dataset -> model lineage (Phase B): - vp snapshot tag v4 trained:run-812 (--remove to undo): free-form tags persisted in the snapshot record, shown in snapshot list/show. - snapshot.py: tag_snapshot/untag_snapshot/find_snapshots_by_tag (a bare "key:" prefix matches any value). --format auto on import (task coverage): - visionpack/formats/detect.py: structural detection — instances-style JSON means COCO (works for a directory with the JSON next to the images too), .txt labels or classes.txt/data.yaml mean YOLO, folder-per-class means ImageFolder; ambiguous layouts raise instead of guessing. - vp import --format now defaults to auto. SDK: new drift(), diff(), tag_snapshot(), untag_snapshot(), snapshots_by_tag(); import_dir defaults to format="auto". Docs: usage.md (auto-detect, lineage, drift sections; limitation removed), json-output.md rows, sdk.md, README, CHANGELOG, ARCHITECTURE roadmap checkboxes. Tests: 15 new (tests/test_drift_lineage.py); 180 total passing. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RKKNgi6ahA6jYnX2CzDEG9
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Security & robustness (from a code audit of the framework):
safe_path_component): names arriving from imported data (folder names,
COCO categories, classes.txt) could contain separators or
..and writeoutside the export directory via
vp export --format imagefolder.image (header claiming absurd dimensions) becomes a per-file ingest
failure instead of aborting the whole import/sync batch.
Performance:
every asset record where sync/import only need the id set.
vp audit (roadmap Phase B — label-health audit):
boxes, edge-pinned and whole-image boxes, aspect-ratio outliers, rare
classes, class imbalance; thresholds via validation.audit in
visionpack.yaml or per-run flags.
Python SDK:
import_dir, sync/plan_sync, validate, audit, stats, splits, snapshots
(with read-only checkout(version) views), export, and the model loop
(load_predictions, evaluate, autolabel, annotation_queue). Mutating
methods take the same project lock as the CLI; returns mirror the
--json envelope shapes.
Docs: new docs/sdk.md, vp audit section in usage.md, json-output table row,
ARCHITECTURE module map + roadmap checkbox, README and CHANGELOG updates.
Tests: 22 new (tests/test_audit.py, tests/test_sdk.py); 165 total passing.
Co-Authored-By: Claude Fable 5 noreply@anthropic.com
Claude-Session: https://claude.ai/code/session_01RKKNgi6ahA6jYnX2CzDEG9